sequencer

Sequencer is a deep learning tool designed for image classification, employing a novel architecture that leverages long short-term memory (LSTM) networks instead of traditional self-attention mechanisms found in Vision Transformers. Its unique Sequencer2D model enhances performance by decomposing LSTM layers into vertical and horizontal configurations, achieving competitive top-1 accuracy of 84.6% on ImageNet-1K with a manageable parameter count. Notable features include good transferability and robust resolution adaptability, making it a compelling alternative in the evolving landscape of computer vision architectures.